Neural Network Models of Language

语言的神经网络模型

• 批准号: 6902613
• 负责人: RALPH Edward HOFFMAN
• 金额: $ 22.07万
• 依托单位: YALE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2003
• 资助国家: 美国
• 起止时间: 2003-07-01 至 2008-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/6902613
• 关键词: behavioral /social science research tag; clinical research; computational neuroscience; human middle age (35-64); human subject; language; neural information processing; neuropsychology; psychopathology; schizophrenia; short term memory; young adult human (21-34)

DESCRIPTION (provided by applicant): Schizophrenia is characterized by alterations of language and inferential processes. In spite of extensive research, core mechanisms of these disturbances remain uncertain. The overall objective of this RO1 proposal is to use DISCERN, a neural network simulation of natural language processing (Miikkulainen & Dyer 1991; Miikkulainen 1993, 1998), to investigate the mechanism(s) of language-based disturbances in schizophrenia. DISCERN learns stories, utilizes inferential processes, replies to questions, and produces coherent, multi-sentence narrative paraphrases of episodic memories. To enhance applicability of DISCERN as a model of human narrative language production, a larger corpus of stories will be learned that incorporates emotion-coding and self-reference. Simulations will be conducted to determine if disrupted function in different neural modules of DISCERN can produce three core language-based illness manifestations of schizophrenia -- (I) positive thought disorder (such as derailment and illogicality), (II) negative thought disorder (reduced language outputs), and (III) delusions of the idee fixe type. DISCERN will be used to compare and contrast effects of excessive noise versus reduced network connectivity when applied to semantic and working memory modules. Both types of "lesions" have been postulated to play an important role in the pathophysiology of schizophrenia. Noise-induced lesions are predicted to produce word selection errors and curtail language output -- but not to produce positive thought disorder or delusions. In contrast, connectivity loss, when applied to story processing modules, is predicted to simulate all three disturbances, i.e., derailment and curtailment of language outputs as well as production of "fixed" narratives that simulate delusions. A parallel, pilot study of normal subjects and patients with schizophrenia will assess narrative recall of episodic memory. These behavioral data will be used to test and refine models of normal and schizophrenic language production. These findings will significantly advance our understanding of illness mechanisms in schizophrenia and direct future research aimed at developing more selective treatments that reverse these abnormalities.

描述（由申请人提供）：精神分裂症的特点是语言和推理过程的改变。尽管进行了广泛的研究，这些干扰的核心机制仍然不确定。 RO1 提案的总体目标是使用 DISCERN（一种自然语言处理的神经网络模拟）（Miikkulainen & Dyer 1991；Miikkulainen 1993、1998）来研究精神分裂症中基于语言的障碍的机制。 DISCERN 学习故事，利用推理过程，回答问题，并产生连贯的、多句子的情景记忆叙述释义。为了增强 DISCERN 作为人类叙事语言生成模型的适用性，我们将学习包含情感编码和自我参照的更大的故事语料库。将通过模拟来确定DISCERN不同神经模块的功能被破坏是否会产生精神分裂症的三种基于语言的核心疾病表现——（I）积极思维障碍（如出轨​​和不合逻辑），（II）消极思维障碍（减少语言输出），以及（III）idee fixe 类型的妄想。当应用于语义和工作记忆模块时，DISCERN 将用于比较和对比过多噪声与网络连接减少的影响。两种类型的“病变”都被认为在精神分裂症的病理生理学中发挥着重要作用。噪音引起的损伤预计会产生词语选择错误并减少语言输出，但不会产生积极的思维障碍或妄想。相比之下，当应用于故事处理模块时，连接性损失预计会模拟所有三种干扰，即语言输出的脱轨和缩减以及模拟妄想的“固定”叙事的产生。对正常受试者和精神分裂症患者进行的一项平行试点研究将评估情景记忆的叙事回忆。这些行为数据将用于测试和完善正常和精神分裂症语言产生的模型。这些发现将显着增进我们对精神分裂症疾病机制的理解，并指导未来的研究，旨在开发更有选择性的治疗方法来逆转这些异常。

项目成果

Using computational patients to evaluate illness mechanisms in schizophrenia.

使用计算患者来评估精神分裂症的疾病机制。

• 发表时间: 2011-05-15
• 影响因子: 10.6
• 作者: Hoffman, Ralph E;Grasemann, Uli;Gueorguieva, Ralitza;Quinlan, Donald;Lane, Douglas;Miikkulainen, Risto
• 通讯作者: Miikkulainen, Risto